Targeted Initial Populations for Multiobjective Product Line Optimization

Abstract

Genetic searches often use randomly generated initial populations that maximize genetic diversity to thoroughly sample the design space. While many of these initial configurations perform poorly, the tradeoff between population diversity and solution quality is typically acceptable for small design spaces. However, as the design space grows in complexity, computational savings and initial solution quality become more significant considerations. This paper synthesizes advancements from market-based design and heuristic optimization research to strategically construct “targeted” initial populations capable of reducing computational cost and improving final solution quality. Respondent-level utilities from a discrete choice model are used with a price segmentation strategy to efficiently populate designs in a multiobjective environment where designers can explore trade-offs between competing business objectives. Results from an automobile feature packaging problem demonstrate the effectiveness of this approach, and recommendations toward the extent of price segmentation required and future research efforts are offered.